Fig. 1.

RV. A conditional variance estimate ($\widehat{\text{Var}}[]$) is computed for a feature, X, at every point in the feature space based on the nearby data points. For points of interest (corresponding to proteins, indicated by p and q), a bandwidth parameter h_p is computed based on the density of points (proteins with similar localization patterns) in that region of the space. The RV compares the conditional variance estimate (surface) to the observed variability (bold P and Q). In this example, the observed variability is high for protein p (P is above the surface) and is low for q (Q is below the surface)